WO2023020853A1 - Procédé d'apprentissage pour entraîner un système d'apprentissage machine artificiel pour détecter des dysfonctionnements, et dispositif de véhicule - Google Patents

Procédé d'apprentissage pour entraîner un système d'apprentissage machine artificiel pour détecter des dysfonctionnements, et dispositif de véhicule Download PDF

Info

Publication number
WO2023020853A1
WO2023020853A1 PCT/EP2022/071976 EP2022071976W WO2023020853A1 WO 2023020853 A1 WO2023020853 A1 WO 2023020853A1 EP 2022071976 W EP2022071976 W EP 2022071976W WO 2023020853 A1 WO2023020853 A1 WO 2023020853A1
Authority
WO
WIPO (PCT)
Prior art keywords
machine learning
learning system
sensors
artificial machine
objects
Prior art date
Application number
PCT/EP2022/071976
Other languages
German (de)
English (en)
Inventor
Arno Hinsberger
Original Assignee
Zf Friedrichshafen Ag
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zf Friedrichshafen Ag filed Critical Zf Friedrichshafen Ag
Publication of WO2023020853A1 publication Critical patent/WO2023020853A1/fr

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M17/00—Testing of vehicles
    • G01M17/007—Wheeled or endless-tracked vehicles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the invention relates to a training method for training an artificial machine learning system for detecting malfunctions of a sensor during operation of a vehicle with multiple sensors.
  • the object is achieved by a training method for training an artificial machine learning system for detecting malfunctions of a sensor with the features of claim 1 and a vehicle device with the features of claim 8.
  • the object is achieved by a training method for training an artificial machine learning system for detecting malfunctions of a sensor during operation of a vehicle with multiple sensors, comprising the steps:
  • An object can be understood to mean any coherent set of attributes or properties of the time curves.
  • Objects can be, for example, vehicles, pedestrians, tracks or traffic signs. For example, this can also be understood as the description of continuous, stationary structures / or the static free space.
  • the time profiles of the objects can be referred to as the path or track of the object, for example. Such a time profile is composed, for example, of the object position, both longitudinally and laterally, and the object speed.
  • Objects are thus specified by their object attributes, which can be evaluated by an artificial machine learning system (e.g. the recognized length, shape).
  • Object parameters are, for example, speed or orientation.
  • a training method for training an artificial machine learning system for detecting malfunctions of a sensor during the operation of a vehicle with multiple sensors is specified.
  • This artificial machine learning system continuously learns the temporal object behavior of the individual sensors. For this purpose, the artificial machine learning system is stimulated with selected objects or their object attributes of the individual sensors in a training phase and the natural object behavior including all continuous and discontinuous deviations is thus trained.
  • the same data can be used for training the artificial machine learning system as for the initial proof of robustness before the start of operation, so that it is ensured that the temporal object behavior with which the artificial machine learning system is trained is not safety-critical.
  • a stable artificial machine learning system can be generated with which unknown temporal object behavior in relation to a sensorically relevant situation can be recognized.
  • the artificial machine learning system is trained for each sensor with the respective object and its temporal behavior.
  • a temporal pattern of the selected objects or object attributes is thus created for each sensor.
  • the measurement noise on which every sensor is based is trained, so that these deviations are no longer recognized as critical deviations.
  • the measurement noise is stronger the further the distance to the object is.
  • the artificial machine learning system trained in this way takes into account properties of the respective sensors, such as the limited resolution of the individual sensors.
  • An artificial machine learning system trained in this way can also take into account the inevitably either continuous and/or sporadic errors that are caused by the sensors in each individual processing step for the individual object attributes.
  • such a trained artificial machine learning system in particular models the natural but non-critical faulty behavior of the individual objects and their object attributes and temporal progressions over time.
  • the machine learning system trained in this way, real deviations in the temporal object behavior and in the object itself can be specifically detected without the risk that actually safety-critical sensor events are overlooked in the mass of system-related (measurement noise) deviations.
  • the non-critical deviations of the temporal object behavior of an object can also be provided as training data.
  • these deviations can, for example, be recognized by the artificial machine learning system itself, for example if a deviation recognized by the artificial machine learning system occurs in a manual evaluation process was classified as non-critical.
  • Object attributes with deviations for example, are used as training data and the "naturally" occurring deviations are thus trained.
  • the artificial machine learning system also learns, for example, continuously deviating object behavior with individual sensors.
  • the at least one artificial machine learning system can be trained individually for each sensor.
  • a separate artificial machine learning system is provided for each sensor, and the separate artificial machine learning system is trained for each sensor, with each individual artificial machine learning system being continuously trained using known objects and the respective object behavior in order to create a pattern of the temporal object behavior as well as the non-critical deviations of the individual objects.
  • each artificial machine learning system also learns, for example, continuously deviating object behavior with individual sensors.
  • two artificial machine learning systems of different types are used for at least two sensors based on different measurement principles. These can be matched to the respective sensors, such as an optical sensor (camera) or radar sensor/lidar sensor.
  • an optical sensor camera
  • radar sensor/lidar sensor for example, as artificial machine learning systems, artificial neural Networks with different hidden layers, different types of artificial neural networks or for the optical sensors Generative Adversarial Networks (GAN) or Convolutional Neural Networks (CNN) can be used.
  • GAN Generative Adversarial Networks
  • CNN Convolutional Neural Networks
  • a number of artificial machine learning systems are provided, with one artificial machine learning system each being trained with the non-critical deviation from individual existing object parameters in order to learn a pattern of the object behavior over time and the non-critical deviations of the individual objects.
  • a number of artificial machine learning systems can be provided, with one artificial machine learning system each being trained with individual object attributes in order to learn a pattern of the temporal object behavior as well as the non-critical deviations of the individual objects.
  • a deviation from an object parameter such as the deviation from the speed for all objects, is used as training data for each artificial machine learning system in order to learn the deviating object behavior of the individual sensors.
  • individual object attributes can also be used as training data. This means that each two artificial machine learning systems receive different object parameters or object attributes for training.
  • a vehicle device for detecting malfunctions of a sensor during operation of a vehicle comprising a plurality of sensors for recording objects in the environment and their chronological object course as respective input data, and a processor unit having at least one trained artificial machine learning system as described above , wherein the processor unit is designed to do so is to apply the at least one trained artificial machine learning system at least to the input data, and wherein the processor unit is also designed to check for the presence of an unknown deviation in a temporal object behavior of an object detected by the sensors in the environmental data during operation of the vehicle using the trained artificial machine learning system for each of the sensors.
  • the vehicle is designed in particular as an autonomously driving vehicle.
  • the vehicle is part of an autonomously driving fleet.
  • the vehicle device is designed to generate a trigger signal upon detection of an unknown deviation in one of the sensors.
  • the trained artificial machine learning system is stimulated with selected objects of the individual sensors during the operational phase in the vehicle. If the observed temporal pattern of the objects deviates from what has been learned, a trigger signal is triggered. As a trigger signal for potentially safety-critical sensor scenarios, only the temporal object patterns that do not match the trained pattern are used during the operative operating phase. The vehicle device designed in this way can now efficiently detect potentially critical sensor errors.
  • the vehicle device is designed to compare the object behavior detected over time by the various sensors and to generate a trigger signal when at least one of the sensors detects persistently inconsistent object behavior over time.
  • inconsistent behavior of the objects with different sensors can also be used as a trigger for potentially safety-critical sensor scenarios.
  • a memory can be provided, with the vehicle device being designed to store the associated, to store objects and their history of objects over time in memory, leading to the unknown deviation.
  • the vehicle device being designed to store the associated, to store objects and their history of objects over time in memory, leading to the unknown deviation.
  • a transmission device which is designed to transmit the associated objects which lead to the unknown deviation and their temporal course of the objects.
  • the data can simply be sent to an external analysis unit which, for example, has more computing capacity available. This allows a more detailed investigation to take place.
  • the vehicle device is designed to continuously feed the continuous sensor signals generated by the sensors to the trained artificial machine learning system. This means that the sensor data is fed to the artificial machine learning system during the entire operation of the vehicle. This allows sensor errors to be detected quickly.
  • the vehicle device is designed to classify the object and its object history into a critical deviation and a non-critical deviation when recognizing an unknown deviation of an object and its temporal object history, and the object and its temporal object history in the case of a non-critical deviation as further training data for the trained artificial machine learning system.
  • the data obtained by trigger signals are analyzed. If the object/object profiles is not critical to safety, the artificial machine learning system can be expanded to include the object pattern and installed on the processor unit that triggered the trigger in the vehicle. This further reduces the number of false positives over time. Further properties and advantages of the present invention emerge from the following description with reference to the enclosed figures. It shows schematically:
  • FIG. 1 shows a training method according to the invention for training an artificial machine learning system for detecting malfunctions in a first embodiment
  • FIG. 2 shows a training method according to the invention for training an artificial machine learning system for detecting malfunctions in a second embodiment
  • FIG. 3 shows a training method according to the invention for training an artificial machine learning system for detecting malfunctions in a third embodiment
  • FIG. 1 shows a training method according to the invention for training an artificial machine learning system for detecting malfunctions of a sensor 2a 2n (FIG. 4).
  • a preliminary step SO which is not part of the training method
  • data are provided by means of which the robustness of the vehicle device 1 (FIG. 4) in relation to the operational mode is quantitatively verified.
  • This data is generated, for example, by simulating or recording real traffic.
  • An object can be understood as any coherent set of attributes or properties of the courses. Objects can be vehicles, pedestrians, tracks or traffic signs. For example, this can also be understood as the description of continuous, stationary structures / or the static free space.
  • the temporal progression of the objects can be, for example, the web or track of the object.
  • Such a time profile is composed, for example, of the object position, both longitudinally and laterally, and the object speed.
  • Objects are thus specified by their object attributes, which are usable by an artificial machine learning system (e.g. recognized length, shape).
  • Object parameters are, for example, speed or orientation.
  • an artificial machine learning system here an artificial neural network, is provided, which is used to detect malfunctions of a sensor 2a,..., 2n (FIG. 4) during the operation of a vehicle 1 (FIG. 4) with a plurality of sensors 2a, ..., 2n (FIG. 4) is to be trained.
  • known objects are provided as coherent object attributes and their temporal object behavior as training data.
  • the same data that are used for the robustness test can preferably be used for this purpose.
  • the training of the artificial neural network is therefore based on the data that has to be collected before the operational phase anyway in order to prove the initial robustness of the system. So no additional data is required.
  • a step S2 the temporal object behavior of the individual sensors 2a, . . . , 2n (FIG. 4) is continuously learned on the basis of the data collected for the robustness test.
  • the input layer of an artificial neural network is stimulated with the non-critical objects detected by the individual sensors in a learning phase, and the natural object behavior including all continuous and discontinuous deviations is thus trained.
  • one artificial neural network learns the temporal object behavior of the individual sensors 2a, . . . , 2n (FIG. 4) on the basis of the data collected for the robustness test.
  • the input layer of an artificial neural network with the non-critical objects, which are detected by the individual sensors 2a 2n (FIG.
  • Suitable objects are therefore selected for each sensor 2a, .
  • a temporal pattern of the selected objects is thus created for each sensor 2a, ..2n (FIG. 4). Furthermore, for each sensor 2a .2n
  • FIG. 4 the typical, system-related, continuous and/or sporadic errors that occur in each individual processing step of an individual object are inevitably trained as a non-critical pattern. These errors in the sensors 2a,...2n (FIG 4) inevitably occur simply because of the measurement noise, which increases with increasing distance, as does the limited resolution of the individual sensors 2a,..., 2n (FIG 4) . Because of the measurement noise, a completely consistent behavior of all objects detected by the sensors 2a, . . . , 2n (FIG. 4) over the processing cycles is therefore impossible. This can be caught by the artificial neural network.
  • the natural but non-critical erroneous behavior of the individual objects detected by the sensors 2a, . . . , 2n (FIG. 4) is modeled over time.
  • the non-critical deviations from the temporal object behavior of an object can also be used as training data, so that the artificial neural network learns an individual sensor-related pattern of the temporal object behavior and the non-critical deviations of the individual objects for each of the sensors 2a,...,2n .
  • FIG. 2 shows a further embodiment of a training method according to the invention.
  • step AO in a preliminary step AO that is not part of the training method and is identical to step SO (FIG. 1), data are provided by means of which the robustness of the vehicle device 1 in relation to the operational mode is quantitatively verified. This data is generated, for example, by simulating or recording real traffic.
  • an artificial machine learning system is provided here, an artificial neural network, which is to be trained during operation of a vehicle 1 to detect malfunctions of the respective sensor 2a, ..2n (FIG. 4).
  • a step A2 the temporal object behavior of the individual sensors 2a, . . . , 2n is continuously learned on the basis of the data collected for the robustness test.
  • the input layer of all artificial neural networks is stimulated with the objects of the respective individual sensors 2a, .
  • two artificial neural networks of different types can be used for at least two sensors 2a 2n (FIG. 4) based on different measurement principles, such as image sensors or lidar sensors or radar sensors.
  • GAN Generative Adversarial Network
  • CNN Convolutional Neural Network
  • recurrent or deep forward neural networks can be used for radar or lidar signals.
  • the special features of the sensor signals of the different sensor types can be better addressed.
  • the non-critical deviations from the temporal object behavior of an object can also be used as training data, so that each of the artificial neural networks learns an individual sensor-related pattern of the temporal object behavior and the non-critical deviations of the individual objects.
  • FIG. 3 shows a further embodiment of a training system according to the invention.
  • a rule CO which is not part of the training method and is identical to step SO, data are provided by means of which the robustness of the vehicle device 1 in relation to the operational mode is quantitatively verified. This data is generated, for example, by simulating or recording real traffic.
  • step C1 multiple artificial neural networks are provided, which are to be trained to detect malfunctions of the respective sensor 2a, .. 2n (FIG. 4) while a vehicle 3 is in operation.
  • a step C2 the temporal object behavior of the individual sensors 2a, . . . , 2n (FIG. 4) is continuously learned on the basis of the data collected for the robustness check.
  • the input layer of an artificial neural network is stimulated with a single selected object attribute, which has been detected by the individual sensors 2a, ..., 2n (FIG 4), and so the natural object behavior including all continuous and discontinuous deviations trained.
  • FIG. 4 shows a vehicle 3 with a vehicle device 1 according to the invention
  • Vehicle 3 with such a trained neural network Vehicle 3 with such a trained neural network.
  • the neural network or networks trained in this way are continuously stimulated during the operational operation of the vehicle 3 with objects recorded in the process. If the observed temporal discrepancy pattern of the sensors 2a, . . . 2n deviates from what has been learned, a trigger signal 8 (FIG. 5) is triggered.
  • the vehicle device 1 can be designed to compare the temporally detected object behavior of the sensors 2a, ... 2n with each other and a trigger signal 8 (FIG 5) upon detection of permanent inconsistent temporal object behavior of at least one of the sensors 2a, ... 2n to generate.
  • a trigger signal 8 (FIG. 5) is triggered, the data required for a sensor-related analysis of the scenario is stored persistently on a memory 4 that is carried along in the vehicle 3 . This means that incorrect object behavior results in all the information required for further analysis, ie the objects/object attributes and other environmental data, being stored statically in a memory 4 .
  • This data is then analyzed and evaluated as a non-critical or critical scenario.
  • the relevant neural network for the sensor is trained with this pattern in order to avoid further unnecessary trigger signals 8 .
  • the artificial neural network is continuously trained.
  • the training of the neural network or networks is expanded to include the object pattern that triggered the trigger signal 8 in the vehicle 3 . This further reduces the number of "false positives" over time.
  • 5 shows the training method and the vehicle device 1 in an overview.
  • the vehicle device 1 is first stimulated with raw sensor data 6 for quantitative verification of the robustness of the environment detection against unknown malfunctions (SOTIF Area 3), which were previously recorded in an endurance run with a statistically significant scope.
  • the objects/object attributes and their chronological course as well as the object parameters are known from these raw sensor data 6 and can be used for the training of a machine learning system 7 or the several machine learning systems 7 . Due to the fact that the same data is used for the training of the machine learning system 7 as for the initial proof of robustness before the start of operation, it is assumed that the temporal object behavior is not safety-critical.
  • the temporal object behavior of the individual sensors 2a On the basis of the objects/object attributes collected for the robustness test, the temporal object behavior of the individual sensors 2a, .
  • the machine learning system or systems 7 are stimulated with objects that were generated by the individual sensors 2a, . . . , 2n, and the natural object behavior including all continuous and discontinuous deviations is thus trained.
  • the training of the machine learning system(s) 7 thus models the natural but non-critical faulty behavior of the individual objects over time.
  • the trained machine learning system or systems 7 are stimulated in the vehicle 3 with the selected objects of the individual sensors 2a, .., 2n in the vehicle device 1 during the operational phase. If the observed temporal pattern of the objects deviates from what has been learned, a trigger signal 8 is triggered.
  • the data (objects/object attributes/environment data) which caused the trigger signal 8 to be triggered are then stored in an on-board memory 4 for a sensor-related analysis.
  • the training of the trained machine learning system(s) 7 is expanded to include the object pattern that triggered the trigger signal 8 in the vehicle device 1 . This further reduces the number of false positive candidates over time.
  • An update 5 of the trained machine learning system(s) 7 can then be carried out in the vehicle device 1 .

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

L'invention concerne un procédé d'apprentissage pour entraîner d'un système d'apprentissage machine artificiel (7) pour détecter des dysfonctionnements dans un capteur (2a,..,2n ) pendant le fonctionnement d'un véhicule (3) muni de multiples capteurs (2a,..,2n), comprenant les étapes consistant à : - fournir des objets connus en tant qu'attributs d'objet associés et le comportement d'objet temporel desdits objets en tant que données d'apprentissage, - fournir au moins un système d'apprentissage machine artificiel (7) ; - entraîner le système d'apprentissage machine artificiel (7) pour chacun des capteurs (2a..,2n) en continu sur la base des objets connus ayant le comportement d'objet temporel respectif associé à l'objet, de sorte que le système d'apprentissage machine artificiel (7) enseigne pour chacun des capteurs (2a,..,2n) un modèle lié à un capteur individuel du comportement d'objet temporel des objets individuels et des variations non critiques des objets individuels. La présente invention concerne également un dispositif de véhicule.
PCT/EP2022/071976 2021-08-17 2022-08-04 Procédé d'apprentissage pour entraîner un système d'apprentissage machine artificiel pour détecter des dysfonctionnements, et dispositif de véhicule WO2023020853A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
DE102021208988.7 2021-08-17
DE102021208988.7A DE102021208988A1 (de) 2021-08-17 2021-08-17 Trainingsverfahren zum Trainieren eines künstlichen maschinellen Lernsystems zur Erkennung von Fehlfunktionen und Fahrzeugvorrichtung

Publications (1)

Publication Number Publication Date
WO2023020853A1 true WO2023020853A1 (fr) 2023-02-23

Family

ID=83115417

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2022/071976 WO2023020853A1 (fr) 2021-08-17 2022-08-04 Procédé d'apprentissage pour entraîner un système d'apprentissage machine artificiel pour détecter des dysfonctionnements, et dispositif de véhicule

Country Status (2)

Country Link
DE (1) DE102021208988A1 (fr)
WO (1) WO2023020853A1 (fr)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102023115307B3 (de) 2023-06-13 2024-07-25 Audi Aktiengesellschaft Verfahren zum Betreiben eines Fahrzeugs, Verfahren zum Entwickeln einer Einheit für ein Kraftfahrzeug sowie Kraftfahrzeug

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102018206208A1 (de) * 2018-04-23 2019-10-24 Robert Bosch Gmbh Verfahren, Vorrichtung, Erzeugnis und Computerprogramm zum Betreiben eines technischen Systems
DE102019105343A1 (de) * 2019-02-09 2020-08-13 Elmos Semiconductor Aktiengesellschaft Messsystem im Fahrzeug zur Erkennung und Klassifizierung von Objekten im Umfeld des Fahrzeugs mit Hilfe eines Deep-Learning Verfahrens
US20210018590A1 (en) * 2020-09-24 2021-01-21 Intel Corporation Perception system error detection and re-verification
DE102019215902A1 (de) * 2019-10-16 2021-05-20 Robert Bosch Gmbh Verfahren zur Bestimmung eines Gütegrades von Daten-Sätzen von Sensoren

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070126561A1 (en) 2000-09-08 2007-06-07 Automotive Technologies International, Inc. Integrated Keyless Entry System and Vehicle Component Monitoring
DE102018209250A1 (de) 2018-06-11 2019-12-12 Bayerische Motoren Werke Aktiengesellschaft Steuereinrichtung, Verfahren zur Steuerung eines Steuergeräts, computerlesbares Speichermedium und Steuersystem
DE102018131964A1 (de) 2018-12-12 2020-06-18 Bayerische Motoren Werke Aktiengesellschaft Verfahren und Vorrichtung zum Verhindern einer ungewollten Beschleunigung eines Kraftfahrzeugs
US11927502B2 (en) 2019-04-29 2024-03-12 Nvidia Corporation Simulating realistic test data from transformed real-world sensor data for autonomous machine applications
DE102019217393A1 (de) 2019-11-11 2021-05-12 Robert Bosch Gmbh Verfahren zum Betreiben einer autonomen Fahrfunktion eines Fahrzeugs

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102018206208A1 (de) * 2018-04-23 2019-10-24 Robert Bosch Gmbh Verfahren, Vorrichtung, Erzeugnis und Computerprogramm zum Betreiben eines technischen Systems
DE102019105343A1 (de) * 2019-02-09 2020-08-13 Elmos Semiconductor Aktiengesellschaft Messsystem im Fahrzeug zur Erkennung und Klassifizierung von Objekten im Umfeld des Fahrzeugs mit Hilfe eines Deep-Learning Verfahrens
DE102019215902A1 (de) * 2019-10-16 2021-05-20 Robert Bosch Gmbh Verfahren zur Bestimmung eines Gütegrades von Daten-Sätzen von Sensoren
US20210018590A1 (en) * 2020-09-24 2021-01-21 Intel Corporation Perception system error detection and re-verification

Also Published As

Publication number Publication date
DE102021208988A1 (de) 2023-02-23

Similar Documents

Publication Publication Date Title
DE102017217733A1 (de) Prüfen eines neuronalen Netzes
DE102008013366B4 (de) Verfahren zur Bereitstellung von Information für Fahrerassistenzsysteme
DE102016220913A1 (de) Verfahren und Vorrichtung zur Generierung von Testfällen für autonome Fahrzeuge
DE102017214611B4 (de) Verfahren zum Überprüfen eines Reaktionssignals einer Fahrzeugkomponente sowie Überprüfungsvorrichtung und Kraftfahrzeug
WO2021121695A1 (fr) Procédé, appareil et système de détection d'états de fonctionnement anormaux d'un dispositif
DE102017210787A1 (de) Verfahren und Vorrichtung zum Ermitteln von Anomalien in einem Kommunikationsnetzwerk
DE102017219473A1 (de) Verfahren zum vorausschauenden Erkennen eines Ausfalls einer Komponente eines Fahrzeugs, computerlesbares Medium, System, und Fahrzeug umfassend das System
DE102017006260A1 (de) Verfahren zum Bestimmen von Detektionseigenschaften wenigstens eines Umgebungssensors in einem Fahrzeug und Fahrzeug, eingerichtet zur Durchführung eines solchen Verfahrens
WO2019211293A1 (fr) Procédé pour faire fonctionner un système d'aide à la conduite d'un véhicule autonome équipé d'au moins un capteur d'environnement pour détecter un environnement du véhicule autonome, support lisible par ordinateur, système et véhicule
WO2023020853A1 (fr) Procédé d'apprentissage pour entraîner un système d'apprentissage machine artificiel pour détecter des dysfonctionnements, et dispositif de véhicule
WO2016012387A1 (fr) Dispositif et procédé de détection de défaillance et d'intrusion pour un véhicule automobile
EP3945338A2 (fr) Chemin de traitement du signal, dispositif de détection de l'environnement et procédé de validation d'un système de conduite à fonctionnement automatisé
WO2021089591A1 (fr) Procédé d'entraînement d'un réseau de neurones artificiels, programme informatique, support d'enregistrement, dispositif, réseau de neurones artificiels et utilisation du réseau de neurones artificiels
DE102019206720B4 (de) Überwachung eines KI-Moduls einer Fahrfunktion eines Fahrzeugs
DE102018004125A1 (de) Verfahren zum Analysieren eines Objekts in einer Umgebung eines Kraftfahrzeugs
DE102020119954A1 (de) Verfahren zum Erzeugen einer Belegungsgitterkarte für zumindest ein statisches Objekt, Computerprogrammprodukt, computerlesbares Speichermedium sowie Assistenzsystem
DE102017005671A1 (de) Verfahren zur Bestimmung von Fehldetektionen eines Sensorsystems eines Kraftfahrzeugs
DE102020203829A1 (de) Verfahren zum Überwachen eines Fahrzeugsystems zum Erkennen einer Umgebung eines Fahrzeugs
DE102019204269A1 (de) Verfahren und Einrichtung zum Erzeugen eines Statussignals
EP4186775B1 (fr) Procédé et dispositif de détection des propriétés d'un véhicule
DE102017213771A1 (de) Verfahren und Vorrichtung zum Ermitteln von Anomalien in einem Kommunikationsnetzwerk
DE102017219269A1 (de) Klassifizierung mit automatischer Auswahl aussichtsreicher Lerndaten
EP3985565A1 (fr) Procédé et dispositif de vérification d'un système de traitement d'informations basé sur l'ia utilisé lors de la commande partiellement automatisée ou entièrement automatisée d'un véhicule
DE102022000257A1 (de) Verfahren zur Umgebungserfassung für ein Fahrzeug
EP4079007A1 (fr) Test de fiabilité automatisé d'un système de capteur de surveillance monté sur une infrastructure

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22761164

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 22761164

Country of ref document: EP

Kind code of ref document: A1